Huber Principal Component Analysis for large-dimensional factor models

Factor models have been widely used in economics and finance. However, the heavy-tailed nature of macroeconomic and financial data is often neglected in statistical analysis. To address this issue, we propose a robust approach to estimate factor loadings and scores by minimizing the Huber loss function, which is motivated by the equivalence between conventional Principal Component Analysis (PCA) and the constrained least squares method in the factor model. We provide two algorithms that use different penalty forms. The first algorithm involves an element-wise-type Huber loss minimization, solved by an iterative Huber regression algorithm. The second algorithm, which we refer to as Huber PCA, minimizes the ${\ell }_2$-norm-type Huber loss and performs PCA on the weighted sample covariance matrix. We examine the theoretical minimizer of the element-wise Huber loss function and demonstrate that it has the same convergence rate as conventional PCA when the idiosyncratic errors have bounded second moments. We also derive their asymptotic distributions under mild conditions. Moreover, we suggest a consistent model selection criterion that relies on rank minimization to estimate the number of factors robustly. We showcase the benefits of the proposed two algorithms through extensive numerical experiments and a real macroeconomic data example. An R package named “HDRFA” ¹ has been developed to conduct the proposed robust factor analysis.